Red Hat Developers Working Towards A Vendor-Neutral Compute Stack To Take On NVIDIA's CUDA

[Jumbotron]

This would be HUGE! I think most folks are missing the BIGGER point which is CUDA customers desire an alternative. A potentially MUCH CHEAPER alternative to the Nvidia tax. It HAS to be comparably performant to CUDA of course. But I think all the raw parts are there for someone like Red Hat...(particularly NOW with help from IBM)...to pull it all together for an open GPU compute stack. There is an almost UNIVERSALLY recognized desire for an alternative. The market needs and desires one. And it will come. Nvidia is at the point where its further upward innovation trendline is flattening at a faster rate than the upward moving trendline for an open source stack. It will probably be another 5 years for the trendlines to nearly match and another 5 years for the market to adopt such an open source stack. But it's coming.

[finalzone]

They're basically saying they'll work on OpenCL support, and see if they can implement OpenCL on Vulkan to reduce the amount of work they need to do on newer GPUs. .

That is the goal from Khronos after Open 2.2 specification is complete. https://www.khronos.org/news/press/k...ith-spir-v-1.2

[schmidtbag]

Although I'm certainly not complaining about this, the only real benefit I see for this is running pre-built CUDA applications on non-Nvidia hardware. Of course, there are a decent amount of such applications, but as far as I'm concerned, the main benefit of using CUDA over OpenCL is the fine-tuned optimizations for Nvidia hardware, and, direct support from Nvidia. I don't think there's going to be a real performance improvement vs OpenCL.
In other words, although I overall support what this project is trying to accomplish, I think it's more important to encourage devs to use OpenCL instead of CUDA.

[airlied]

Michael the video will be out soon I think, but one thing the video states is this is NOT a Red Hat driven project. The idea of the talk was to get some feedback from upstream Linux on the idea of an upstream Linux developed stack, not on some Red Hat product solution. All the other work done by Karol, Rob and Jerome has a real world goal, my talk is a wouldn't it be nice, what if we could make something upstream like the kernel, so currently it's just me throwing around ideas not a Red Hat driven plan.

[TemplarGR]

This is a great idea. The key is a great implementation, and i have my doubts about that. Still i wish them the best!

[airlied]

Michael the whole opening sentence is untrue, none of this is a long term plan, hiring nouveau Devs has 0 to do with my talk, your article is too speculation heavy, maybe watch the video when it arrives, it might clear up the speculation.

[Michael]

Michael the whole opening sentence is untrue, none of this is a long term plan, hiring nouveau Devs has 0 to do with my talk, your article is too speculation heavy, maybe watch the video when it arrives, it might clear up the speculation.

Thanks for clarification, cleaning it up.

[msroadkill612]

My newb general impression of the gpu compute world is the pervasive view that cudas superiority is set in stone, yet it seems such a young field. Surely its far to early to call.

Business history tells us they certainly WILL lose their dominance & protective moat. Its just a question of when.

CUDA is dominant - fine - I accept that, but their position is nothing like assured after such a short time in such a young and dynamic field.

[airlied]

The stack sounds great, with one _major_ exception: SYCL. There is over 10 years of successful development based on CUDA. CUDA is a proven programming model -- it extends C++ in a minimal fashion and there is a ton of code already written in it. Please don't bother with SYCL, it has already failed to achieve anything significant, and is doomed to fail like OpenCL before. Instead, re-target the CUDA support that is already implemented in Clang to target the proposed stack, circumventing nvptx/ptxas,and this will be a massive success.

The video goes into more detail on why SYCL, but you can't create a standard around CUDA without NVIDIA giving CUDA to a standards body, it kinda limits your choices.

If you don't have a standard, NVIDIA can remove the rug at any point.

[airlied]

My newb general impression of the gpu compute world is the pervasive view that cudas superiority is set in stone, yet it seems such a young field. Surely its far to early to call.

Business history tells us they certainly WILL lose their dominance & protective moat. Its just a question of when.

CUDA is dominant - fine - I accept that, but their position is nothing like assured after such a short time in such a young and dynamic field.

People used to think there was only Windows, only Solaris, those people learn over time.

Dave.